DITlieD

I'm not a parasite anymore.

[DITlieD @DITlieD]

Devin is an underrated harness, less over engineered than hermes, and more useful.

Coming from someone building their own harness for 8+ months.. Devin improved a lot, i remember reverse engineering it 5 months ago and it was 100x worse.

[DITlieD @DITlieD]

@anime_shots Anime ?

[DITlieD @DITlieD]

@AnimeBasee What anime is this ?

[DITlieD @DITlieD]

@Rinne_yt He literally looks down for a second

[DITlieD @DITlieD]

@velvet4venus Is this the Anime ? woh

[DITlieD @DITlieD]

@filicroval Bro lying so hard, Alpha is not AGI bro, that model and animation + UI assets in one prompt ? hah.

[DITlieD @DITlieD]

@TJeparskis @AlexNkDc No its not

[DITlieD @DITlieD]

@thsottiaux @TJeparskis Work harder then, this is 2 sol max sessions, each hit 500k context and used 3-5 luna max sub agents.

This is not the kind of usage i expected considering luna and sol prices.. anthropic 20x was giving me much more usage, 2 500k on opus 5 was barely 1% of the weekly..

[DITlieD @DITlieD]

@TJeparskis Why lie ? This is 2 sol max sessions, each hit 500k context and used 3-5 luna max sub agents, so explain what "half an hour" means for you please.

[DITlieD @DITlieD]

@AI_Illust_000 love her eyes

[DITlieD @DITlieD]

Well, reasoning loops are a known problem, i just turned off the reasoning effort level and i use my own harness to make them think as necessary, it achieves the "todo" style reasoning of Ox Alpha the exact same way, but with real output tokens and actions not "i read X file" without ever reading it

[DITlieD @DITlieD]

@RhysSullivan bro what is this lol vibe coders have nothing useful to do ?

[DITlieD @DITlieD]

@_nexuslink @maria_rcks I mean i agree with her, i prefer to have a clean 4 tool harness + my own stuff on top than a overengineered slop mess

[DITlieD @DITlieD]

huihui-ai/Huihui-Ornith-1.5-9B-abliterated — huihui-ai's standard abliteration (weights orthogonalization, minimal KL).
zaakirio/Ornith-1.5-9B-Uncensored — decensored with the Heretic method (refusal direction removal), plus a GGUF quant (zaakirio/Ornith-1.5-9B-Uncensored-GGUF).
junafinity/Ornith-1.5-9B-uncensored — another uncensored finetune, also with GGUF-8bit and MLX quants.

[DITlieD @DITlieD]

@elder_plinius Oh so your slop workflows caught you.. i dont understand why you are so famous, i guess normies like normies.

[DITlieD @DITlieD]

You know the model only sees whats on his context and generally effort level isnt unless you put it in there + wtf is fowlcode ? that low, 10 out of 100 values are hallucinated or you made smt to tell the model its effort level and that is not working properly, or you are just attention farming.

[DITlieD @DITlieD]

@Da7_Tech Are you using a local proxy to save all data like tokens you use etc so after sometime you can publish how much the 200 usd was worth it in api prices

[DITlieD @DITlieD]

@Da7_Tech @velntaw excited for grok 4.7 1m context + it will beat fable 5 at 5/10 per 1m i think

[DITlieD @DITlieD]

@1camey_ @matt_kaschel @dan0madpro @elder_plinius full self improvement loop ? bro it updates its skills and create useless memory files, .md spaguetti, its lazy because you guys dont care if its good or bad, you see words you dont understand and you think its magic

[DITlieD @DITlieD]

@Chris_Segroves @elder_plinius Anything can, the latency between the machines is real tho